Research On Detection Technology Of Black Tea Fermentation Quality Information Based On Hyperspectral

Fermentation is an important process that affects the quality of black tea,the degree of fermentation directly determines the quality and flavor of black tea.At present,the judgment of the quality of black tea fermentation at home and abroad mainly depends on the experience of the tea maker,which is highly subjective and easily leads to inconsistent quality of the finished tea.In order to accurately determine the quality of black tea fermentation,it is necessary to use equipment and continuous physical and chemical testing of key quality indicators.Obviously,this method can't quickly perceive the quality information of black tea fermentation and can't meet the needs of black tea industrial production.Therefore,it is particularly important to seek an effective rapid non-destructive testing method to distinguish the different fermentation quality of black tea.This study is mainly aimed at the fermentation process of Congou,taking samples under different fermentation timing as the object,using hyperspectral detection technology and chemometrics methods to establish a fermentation degree discrimination model and a quantitative prediction model of key quality components under the fermentation time sequence,and the visualization of key quality components of fermented leaves was realized by the optimized model,which provided theoretical methods and scientific basis for advancing the intelligent processing of black tea.The main research contents of this paper were as follows:A method for judging the fermentation quality of Congou based on hyperspectral.First,the hyperspectral imager(400?1000nm)was used to collect the hyperspectral data of Congou fermentation samples,and according to the on-site production information such as temperature,tea tenderness,withering condition,rolling process,fermentation leaf color and aroma,etc.,six different fermentations samples under the time series were divided into three categories according to the degree of fermentation(light fermentation,moderate fermentation,excessive fermentation).In order to eliminate the influence of scattering phenomenon on the spectral data when collecting hyperspectral information,the preprocessing algorithms was used to preprocess the original spectra to reduce the dimensionality of hyperspectral data,The variable selection algorithm was used to extract characteristic wavelengths and PCA was used to optimize the data.KNN,RF,ELM,SPA-KNN,SPA-RF and SPA-ELM discriminant models were established.Finally,the advantages and disadvantages of different models were analyzed,and the models used in actual production were determined.It is hoped that the research foundation will be laid for the development of on-line determination equipment for black tea fermentation.Quantitative prediction and visualization of key physical and chemical components in black tea fermentation using hyperspectral imaging.The collected hyperspectral data of the sample corresponds to the analysis value of the physical and chemical detection of the sample under different time series.Principal component analysis(PCA)was used to optimize the dimensionality of the HSI data.The relationship between the position of the stacked fermentation leaves and the key components was discussed,and the influence of different preprocessing,variable screening and intelligent algorithms on the performance of the model was analyzed.Different quantitative analysis models of key physical and chemical components in fermentation(theaflavins,thearubigins,theafuscins,caffeine,phenol-ammonia ratio,catechins and soluble sugars)have been established,and the best prediction model of each key physical and chemical composition has been determined.And according to the prediction result of the best prediction model corresponding to the background position information,the visual images of the quality components of the piled fermented leaves at different times were obtained after the pseudo-color processing.The validation of Congou black tea fermentation degree discrimination model and key quality component prediction model.In the black tea fermentation experiment,in order to make the model more accurate and comprehensive to predict the fermentation information of black tea under different time series,the stacked fermented leaves were divided into three layers,each with a thickness of 5 cm.When collecting hyperspectral information,the interval was 1 hour,and each layer of fermented leaves was collected once.In order to make the built model reflect the overall information of black tea fermentation,the hyperspectral data of the upper,middle,and lower layers of black tea at different fermentation timings were extracted,and the black tea fermentation degree discrimination model and the key quality component quantitative prediction model in the fermentation were established respectively.However,in actual production,when using hyperspectral collection to collect information about the fermentation of black tea,only the information on the surface of the fermented leaf can be collected.In order to test the accuracy of the discriminant model and the quantitative prediction model in actual production,the hyperspectral data of the upper layer at different moments of black tea fermentation was extracted,and the data was input into the discriminant model and quantitative prediction.The discriminant effect and prediction result of the model were observed,and the appropriate fermentation time and the physical and chemical test results of key quality components were compared.